learning signals within sensory environments: does host cue … · 2019-03-29 · animal biology,...

Animal Biology Vol 56 No 2 pp 173-192 (2006) Koninklijke Brill NV Leiden 2006Also available online - wwwbrillnl

Learning signals within sensory environmentsDoes host cue learning in butterflies dependon background

EMILIE C SNELL-ROOD lowast DANIEL R PAPAJ

Department of Ecology and Evolutionary Biology University of Arizona BioSciences West Rm 3101041 East Lowell St Tucson AZ USA 85721

AbstractmdashInsects must detect and interpret stimuli embedded in a sensory environment of competingstimuli While sensory environments vary in time and space individuals may be able to learn localbackground characteristics facilitating perceptual learning This study on host search in butterfliesexamines the following questions in an ecologically relevant context i) does cue learning dependon the sensory environment in which learning occurs and ii) are background characteristics learnedsuch that performance on novel tasks in the same sensory environment is facilitated Females of Battusphilenor (Papilionidae Lepidoptera) were trained to different coloured and shaped oviposition targetsagainst different background colours Individuals trained to colours on a brown background but testedon a green background performed significantly worse than control individuals which were trainedto the same colours but on a green background Females pre-trained to discriminate green targetsfrom red targets on a green background colour performed significantly better in a novel task (shapelearning) involving green shapes on a green background than did individuals trained to discriminatethe same colours on a brown background These two results were unique to particular cue-backgroundcombinations in particular cryptic conditions Taken together our results suggest that cue learningdepends on an insectrsquos sensory environment and that learning characteristics of local backgroundsmay confer benefits to habitat-faithful individuals

Keywords butterflies learning background Battus philenor context-dependency crypticity fil-ters sensory noise signal detection

INTRODUCTION

Organisms do not detect signals and cues in a sensory vacuum such stimuli oc-cur against a background of competing stimuli The background or sensory en-vironment includes all stimuli within an environment including irrelevant stimuli

lowastCorresponding author e-mail emiliesemailarizonaedu

174 EC Snell-Rood DR Papaj

termed lsquonoisersquo which are similar to and thereby often obscure stimuli of inter-est Sensory environments are moreover inherently variable over time and spaceFor example different habitats are characterised by visual environments that varyin terms of the spectral composition of ambient light and the reflectance proper-ties of vegetation and substrates (Endler 1993 Marchetti 1993 Endler and Theacutery1996 Leal and Fleishman 2004) Ambient sound levels also vary significantly be-tween habitat types some habitats being consistently noisy over certain frequencyranges (eg Morton 1975 Ryan and Brenowitz 1985 Slabbekoorn and Smith2002) Acoustic noise due to sound reverberation varies between seasons in de-ciduous forests (Naguib 2003) and between habitats (Richards and Wiley 1980)Phytophagous insects must discriminate host plants against olfactory cues from non-host plants olfactory background characteristics depend on factors such as non-hostplant species present vegetation structure weather conditions and in an agricul-tural context the diversity and spacing of crops (reviewed by Visser 1986)

Species often adapt to cope with characteristics specific to particular sensoryenvironments For example signals used in communication evolve to travel ef-ficiently through particular habitats (eg Ryan and Brenowitz 1985 Marchetti1993 Endler and Theacutery 1996 Slabbekoorn and Smith 2002 Cunningham et al2002 reviewed by Wiley 1994) Sensory systems evolve not only to perceive rele-vant information (Wehner 1987) but also to filter out irrelevant background infor-mation (reviewed by Endler 1992)

Additionally individuals may dynamically adjust signals to spatiotemporal varia-tion in sensory environments Such individual level responses have been best stud-ied from the perspective of the signaller in communication systems When signalsare obscured by the background signals tend to increase in intensity or in numberof repetitive elements (Cynx et al 1998 Lengagne et al 1999 Brumm and Todt2002 Brumm 2004) Alternatively signallers may opt to signal later when back-ground conditions change (eg delays in communication due to weather conditions(Lengagne and Slater 2002) or time of day (van Staaden and Roumlmer 1997))

Individual-level plasticity in response to sensory environments has been muchless studied from the standpoint of a signal receiver However a receiver may alsorespond to variation in the sensory environment whether with respect to signalsfrom conspecifics or cues from predators or prey Receivers could cope withvariability in sensory environments in several non-mutually exclusive ways Forinstance signal detection theory predicts that receivers may adjust their thresholdof response to a stimulus of interest so as to simultaneously maximise responsesto the relevant stimulus and minimise responses to similar but irrelevant stimuli(so-called lsquofalse alarmsrsquo) A signal detection framework (cf Wiley 1994) hasrecently been applied with success to analysis of learning of floral cues by foragingbumblebees (Lynn Cnaani and Papaj 2005)

Alternatively individuals experiencing a particular sensory environment maylearn through a process of perceptual learning to ignore features of the environmentand thus improve subsequent performance (Watanabe et al 2001) Perceptual

Learning sensory environments 175

learning long studied in psychology (reviewed by Sathian 1998 Goldstone 1998)has been documented in many non-human animals for instance in the formation ofsearch images during search for cryptic prey (Pietrewicz and Kamil 1979 Plaistedand Mackintosh 1995 Langley 1996 Goulson 2000 Bond and Kamil 2002) andin learning of olfactory cues during search for host plants (reviewed by Visser1986 reviewed by Papaj and Prokopy 1989) Theory and experimental evidencefrom cognitive science and psychology predict that perceptual learning involvesnot only learning templates for the cues but also learning to ignore aspects of thesensory environment (Vaina et al 1995 Dosher and Lu 1998 2000 Sigala andLogothetis 2002 Gold et al 2004 Yang and Maunsell 2004) However studies ofsuch lsquobackground learningrsquo in an ecological context are rare

This study addresses the hypothesis that in an ecological context insects learnnot only cues associated with rewards but also the means by which to extract thosecues from the sensory environment We predicted that i) the efficacy of response toa given cue in a given sensory environment will depend on the sensory environmentin which that cue was learned and ii) characteristics of particular backgroundsare learned and transferred to learning of novel tasks in that sensory environmentOur predictions were tested in the context of visual cue learning by host-searchingbutterflies where the lsquosensory environmentrsquo is the visual background surroundingthe host cues

METHODS

Study system

The pipevine swallowtail butterfly Battus philenor L is a papilionid speciescommon in North America that specialises on Aristolochia species over its entirerange In southern Arizona the butterfly uses a single host species Aristolochiawatsoni Wooton amp Standley The Battus philenor-Aristolochia watsoni systemis a particularly appropriate study system for the questions asked here beingcharacterised by considerable variation both in the host resource and the vegetativebackground against which the host occurs In southern Arizona B philenor is activebetween March and September including a dry pre-monsoon period (March-June)and a rainy monsoon season (July-September) Throughout the year the smallhighly recumbent host plants vary markedly in leaf colour ranging from a darkred to a bright green some plants are consistently one colour while others switchcolour and still others consist of mixtures of red and green foliage All forms canoccur at a given site with green forms always rare but becoming more common afterthe monsoon The vegetative background in the mesquite-grassland habitat consistslargely of grasses and herbaceous plants and changes visually from brown andyellow before the monsoon to green after the monsoon Additionally host plantsin washes commonly appear against brown backgrounds independent of seasonadding a component of spatial heterogeneity in background Host-searching females


frequently land on non-host plants in the vicinity of host plants suggesting that thevegetative background against which hosts are found might influence host finding

Battus philenor readily learns oviposition cues including colour and someaspects of shape (Papaj 1986 Allard and Papaj 1996 Weiss and Papaj 2003)and performs well under controlled laboratory conditions Females have beendemonstrated to use colour as a host-finding cue in the field and to learn red or greenreadily under laboratory conditions (Weiss and Papaj 2003 D Papaj unpubl)

Study subjects were obtained from a laboratory colony Larvae were reared onfresh Aristolochia fimbriata leaves replaced daily (1311 LD photoperiod 23C)Adults were transferred to a large flight cage (2 times 2 times 2 m) and hand-fed one tothree times daily on a 15 honey solution Four 500 W halogen lights provided heatand light for courtship for several hours per day Mated females were individuallynumbered on their wings with a gold paint pen Except during oviposition trainingmated females were kept naiumlve in terms of host experience by placement in holdingcages (025 times 025 times 025 m) within the flight cage Training sessions lasted 15-4 h depending on how long it took females to reach training criteria Testing andshape training sessions occurred between 2 and 48 h following training and lasted2-4 h Females used in the experiment were fed before and after training or testingsessions In general females were completely host-naiumlve prior to training howeversome females were provided limited access to a host plant a live Aristolochiawatsoni so as to relieve egg load In ANOVAs including treatment group priorexperience with a host plant had no effect on training performance (F128 = 086P = 036) or test performance (F128 = 016 P = 069)

Oviposition learning

Butterflies were tested on an array of 16 host plant models hereafter referredto as lsquotargetsrsquo arranged in a Cartesian grid with targets spaced 20 cm apartOviposition targets were constructed from paper and consisted of six rectangularlsquoleavesrsquo projecting radially out from an inverted plastic pipette tip Targets were6 cm diam Each oviposition target was placed in the centre of a 20 cm2 paper squareof a certain background colour (green or brown) A square of this size ensuredthat the background was visible to a butterfly from behind the target at an angle ofapproach as low as 25 Oviposition targets were red green or blue Target colourwas generated by printing on white inkjet paper (waterproof lsquoNational GeographicAdventure paperrsquo Teslin) from an Epson Stylus C80 inkjet printer using Durabritebrand inks

Red and green targets were spectrally matched to natural variation in host plantcolour using an Ocean Optics S2000 spectrophotometer Blue was an arbitrarycolour that was similar in peak intensity to the green target In Experiment 1probably the most cryptic treatment group consisted of green targets against a greenbackground as the hue (wavelength of peak intensity) of both the target and thebackground is closely matched In Experiment 2 the green background colour wasdarkened so that the green target was less cryptic against the green background


Figure 1 Spectral reflectance of targets and backgrounds Each graph shows the reflectance (y axis)for each wavelength (x-axis nm) The top graph includes each target used (G = green R = redBL = blue) the middle and bottom graph show the backgrounds used in Experiments 1 and 2respectively (G = green BR = brown)

(fig 1) making the treatment groups of comparable crypticity and to encourageshape learning a more difficult task than colour learning

During training each target had a central cotton wick wet either with water plus150 microl of Aristolochia fimbriata extract (in the training mode S+) or with watertinted to the same orange colour as the extract (in the neutral mode S0) The extractwas prepared by blending 385 g fresh A fimbriata leaves in 675 ml boiling ethanolThe blended solution was filtered under vacuum and ethanol removed under vacuumat 40C until the extract was concentrated to 400 ml This resulted in a concentrationof 1 g of Aristolochia foliage per ml solvent or 1 g leaf equivalent (= 1 gle)The resulting mostly aqueous extract was centrifuged to remove chlorophyll asa particulate The decanted solution was stored in sealed glass containers at minus4CWicks were changed daily


Rewarding and neutral targets were stored separately to prevent cross-contami-nation In both experiments position data were recorded to test for spatial locationbiases (none were found)

Learning assays

Training was initiated either by placing females directly on an S+ target or waitingfor a female to begin host searching spontaneously Direct placement generallyelicited a highly stereotyped oviposition search behaviour characterised by a slowfluttering flight frequent turns and periodic landings on targets While number ofplacements was not strictly controlled varying with individual lifetime and otherfactors beyond our control the number of placements per individual did not differamong the four training groups (F342 = 163 P = 019)

Once a female was in oviposition search mode each landing on a target bythat female was recorded with reference to target number and colour If the targetwas probed (for nectar) the landing was not counted and the individual wasimmediately fed If the individual landed on the target and basked the landing wasalso not counted Landings on the background in proximity to targets were recordedSuccessive landings on the same target were recorded as separate landings only ifbetween landings the individual left the square with the target of interest and pausedin flight over other targets

During test phases individuals were induced to search for hosts by placing themon a wick (separate from any targets) wetted with Aristolochia extract andor placingthem directly on a neutral target During all observations a wick wetted with waterand 1 ml crude Aristolochia extract was placed nearby such that volatiles were likelyto stimulate females to search

Data analysis

To obtain accurate estimates of initial errors and learned responses we used logisticregression as a statistical model of change in behaviour due to learning Logisticregression was applied to the landing data for each individual yielding regressionswith P values of 010 or less for 20 of 44 individuals trained for at least 20 landings(Exp 1) and nine of 23 trained for at least 50 landings (Exp 2) Using the equationfor logistic regression (Eq 1) and the output estimates for the regression parameters(a b) the probability of correct landings (y) was calculated for each individualboth for initial probability of error (x = 0) referred to as lsquoinitial errorrsquo and finalprobability of error (x = total landings for that individual) referred to as lsquofinalerrorrsquo (see fig 2)

y = 1

1 + eminus(a+bx)(1)

lsquoTraining performancersquo was defined as the difference between final error and initialerror (fig 2) lsquotest performancersquo was defined as the difference between the test error


Figure 2 A representative learning curve during training Points represent individual landings oneither the correct rewarding target (Y = 1) or the incorrect non-rewarding target (Y = 0) Alogistic regression is fit to the data to represent the decrease in probability of error over time Twoparameters from this regression the initial error (error at X = 0 landings) and the final error (errorat X = final landing) are used in the analysis This individual was trained to a green target against agreen background

(percentage of incorrect landings throughout the test) and the final error duringtesting (as calculated from logistic regression) All analyses involving proportions(eg initial errors) were arcsine square root transformed prior to analyses

EXPERIMENT 1 ARE CUES LEARNED INDEPENDENTLYOF THE BACKGROUND

Experimental design

To determine if cues are learned independently of background individuals weretrained either to a red or a green target type against either a green or brown back-ground yielding four training groups i) GreenGreen (S+Background) ii) GreenBrown iii) RedGreen or iv) RedBrown (see fig 4) Individuals from each train-ing group were tested against either a green or brown background yielding a totalof eight groups of butterflies At least five butterflies were used in each of the eightgroups The four groups consisting of individuals trained and tested against the samebackground were considered control groups The other four groups consisting of in-dividuals trained on one background but tested against the other background wereconsidered treatment groups During the testing session test targets had wicks wetwith orange coloured water


Learning criteria

Individuals were trained for at least 20 landings on either target (mean = 6917(SD = 4733) N = 45 range = 20-234) Total landings did not differ among thefour training groups (F341 = 096 P = 042) or eight test groups (F726 = 070P = 067) Training lasted for at least 20 landings until performance improvedby at least one landing between the first and final ten landings Following traininglogistic regression was used to confirm that trained individuals actually improvedin performance overall If the lsquotraining performancersquo the difference between thefinal error and the initial error was at least 005 the individual was includedin the analysis This protocol eliminated individuals that did not improve theirperformance during the training session and reduced sample sizes slightly betweengroups 1C 1T 2C 2T 3C 3T 4C 4T N = 5 5 4 4 4 3 4 5 Usingthis criterion individuals included in the analysis reduced their error rate by anaverage of 31 percentage points (SD = 19 N = 44 range = 51-699 percentagepoints)

RESULTS (1)

Initial error and learning during training

Butterflies learned over the course of training to land preferentially on the rewardingtarget Data from a typical training session are shown in figure 2 with a logisticregression fit to the landings to describe changes in the probability of error Amongall butterflies trained the difference between initial and final error was significantlygreater than zero indicating that the rewarding colour was learned (t50 = 867P lt 0001) Training performance (= the difference between initial and finalerror) was related to background colour with stronger performance against brownbackgrounds (fig 3 background parameter F140 = 543 P = 003) performancewas not related to the S+ colour or to an interaction between target and backgroundcolour

The butterfliesrsquo initial error standardised with reference to a green target (ie1 = all green landings 0 = all red landings) was affected by the background(background parameter F126 = 561 P = 002 S+ and interaction NS) thechance of choosing green being significantly higher against a brown than a greenbackground (fig 3) There was no difference among the eight test groups (F726 =149 P = 022) or between control and treatment groups in initial error forgreen There was no difference between control and treatment groups in trainingperformance except for a marginal difference in the red on brown group (P = 008)where the treatment group had higher training performance than the control group(fig 3) however this difference was in a direction opposite that which might biasresults


Figure 3 Summary of training performance The Y-axis represents the initial (black bars) andfinal (grey bars) errors with respect to the green target While statistical tests were performed ontransformed values untransformed data are shown error bars represent standard error All individualsstart at comparable initial errors (slightly biased towards green) and diverge as they learn tochoose only the green or only the red targets Each column represents a different training group ofbutterflies as labelled above light and dark grey backgrounds represent green and brown backgroundsrespectively lsquoGrsquo and lsquoRrsquo represent green and red targets respectively dark circles signify therewarding target (with host plant extract)

Effects of background on memory

The difference between an individualrsquos test error (= total percentage of incorrectlandings during the test) and its final error during training was significantly affectedby treatment in addition to target colour training background and test background(fig 4 overall model F429 = 1118 P lt 00001) As evidenced by a significanttraining background times test background interaction effect (F1 = 933 P = 0005)performance on a given test background depended on treatment group Individualt tests showed that individuals that switched from a brown training background to agreen test background had significantly lower test performance relative to controlsboth for green target training (fig 4 P = 0009) and for red target training (fig 4P = 002)

Performance for individuals trained to green targets deteriorated less betweentraining and test sessions than performance for individuals trained to red targets(S+ parameter effect F1 = 139 P = 0008) Performance for individuals trainedon a green background deteriorated less than performance for individuals trainedon a brown background (F1 = 961 P = 0004) Performance for individualstested on a green background generally exceeded that of individuals tested on abrown background (F1 = 620 P = 002) Remaining interaction terms in ourfully-factorial ANOVA were not statistically significant (F lt 040 P gt 050)


Figure 4 Memory following a change in background The Y-axis represents the difference betweenthe error at the end of the training session (final error) minus the total error during the test (withrespect to the target colour the individual was trained to) The X-axis signifies the training andtreatment regime for a butterfly according to symbols outlined for figure 3 control groups aretrained and tested on the same background while ldquotmtrdquo or treatment groups are trained and testedon different backgrounds Thus a zero value represents complete retention of learned informationbetween training and testing and negative values represent lsquoforgettingrsquo Treatment groups (those thatswitched backgrounds) performed worse than control groups only when training was on a brownbackground and testing on a green background Error bars represent standard error

EXPERIMENT 2 ARE CHARACTERISTICS OF THE BACKGROUNDLEARNED AND APPLIED TO NOVEL TASKS

Experimental design

In this experiment we sought to determine if a female learned characteristics ofthe background during a pre-training colour discrimination task and transferred thislearning to a novel shape discrimination task Female butterflies were first trainedto discriminate rewarding oviposition target from an unrewarding red target For onegroup of females the rewarding target was green for another group the rewardingtarget was blue As in Experiment 1 the targets consisted of a six-pronged radiallysymmetrical shape 6 cm in diam Extracts were applied to rewarding targets asdescribed in Experiment 1 Half of each colour training group was trained againsta green (control) or brown (treatment) background colour (see fig 5) Females ineach target colour times background colour combination were then given training on asecond oviposition task involving discrimination between two novel green shapes


Fig

ure

5Pe

rfor

man

ceon

ano

veld

iscr

imin

atio

nta

sk(s

hape

)fol

low

ing

expe

rien

cew

ithpa

rtic

ular

back

grou

nds

duri

ngco

lour

trai

ning

The

Y-a

xis

repr

esen

tsa

part

icul

arm

easu

reof

perf

orm

ance

duri

ngsh

ape

trai

ning

The

X-a

xis

isas

infig

ure

3ex

cept

lsquoBrsquor

epre

sent

sbl

ueta

rget

san

dsh

apes

inth

elo

wer

box

repr

esen

tth

etw

ota

rget

sus

edin

the

shap

etr

aini

ngT

hele

ft-h

and

grap

hsh

ows

chan

ges

inth

eco

rrec

tch

oice

ofth

ere

war

ding

shap

eov

erth

esh

ape

trai

ning

sess

ion

(ie

di

ffer

ence

betw

een

initi

alan

dfin

aler

rors

see

fig2

)th

usm

ore

posi

tive

valu

esre

pres

ent

impr

ovem

ents

indi

scri

min

atio

ndu

ring

shap

etr

aini

ngT

hem

iddl

egr

aph

show

sla

ndin

gra

tes

onta

rget

sdu

ring

shap

etr

aini

ng(n

ote

log

scal

e)t

hus

mor

epo

sitiv

eva

lues

repr

esen

thi

gher

perf

orm

ance

For

thes

efir

sttw

om

easu

res

ofpe

rfor

man

cet

hegr

oup

with

prev

ious

expe

rien

ceon

the

shap

e-tr

aini

ngba

ckgr

ound

(con

trol

grou

p)pe

rfor

med

bette

rth

anth

egr

oup

wit

hout

such

expe

rien

ce

only

whe

nco

lour

trai

ning

occu

rred

ongr

een

targ

ets

The

righ

t-ha

ndgr

aph

show

sch

ange

sin

abili

tyto

igno

reth

eba

ckgr

ound

betw

een

colo

urtr

aini

ngan

dsh

ape

trai

ning

whi

lein

divi

dual

sle

arne

dto

igno

reth

eba

ckgr

ound

duri

ngco

lour

trai

ning

(see

text

)th

isab

ility

was

notr

etai

ned

insh

ape

trai

ning

asth

edi

ffer

ence

betw

een

initi

alba

ckgr

ound

land

ing

duri

ngco

lour

trai

ning

and

shap

etr

aini

ngw

asle

ssth

anor

equa

lto

zero

(sig

nify

ing

noch

ange

or

anin

crea

sein

back

grou

ndla

ndin

gsbe

twee

ntr

aini

ngse

ssio

nsr

espe

ctiv

ely)


(a rewarding 12-pronged versus an unrewarding three-pronged target of the sameoverall area) against a green background

Several measures of performance during the shape discrimination task were madeFirst we measured the total change in discrimination between the two shapes overat least 20 landings (mean no of landings = 683 (SD = 3355)) Similar to theabove analyses of learning logistic regressions were fitted to the shape trainingdata where landing on a rewarding target was considered a success (Y = 1) andlanding on an unrewarding target was considered an error (Y = 0) Change in shapediscrimination was estimated by subtracting the final error in the shape session fromthe initial error during that session (fig 2)

The second measure of performance was target landing rate (targets per min)during a searching session The time to the nearest second (s) for each landing wasrecorded during shape sessions to facilitate this calculation A searching sessionwas started when a female entered oviposition mode (see description above) andended when the female left the array or was inactive for at least two min Targetlanding rate was calculated as the total number of landings on either rewardingor non-rewarding targets divided by the total time of the session averaged overall searching sessions for each individual with at least two sessions Our measureof target landing rate was skewed towards low values a natural log transformationwas used to normalise the distribution

The third measure of performance was discrimination against background Land-ings on the background as well as landings on targets were recorded during boththe colour and the shape training phases Logistic regressions were computed todetermine if individual butterflies learned to discriminate targets against the back-ground target landings were counted as successes (Y = 1) and background land-ings were counted as errors (Y = 0) We used parameters from logistic regressionto test if background discrimination improved during colour training and if this dis-crimination was retained during shape training If butterflies learn features of thebackground they should learn and remember to ignore (ie not land on) the back-ground

We predicted that during shape training individuals with colour training againsta green background would relative to individuals colour trained against a brownbackground i) learn to discriminate shape faster ii) have higher shape-trainingtarget landing rates and iii) learn to ignore the green background during colourtraining and retain this response during shape training If females learn backgroundcharacteristics dependent on characteristics of the cue these predictions should holdonly when background and cue colours remain the same between discriminationtasks (ie the greengreen colour task and greengreen shape task)

Learning criteria

The protocol for quantifying and analysing learning was identical to that ofExperiment 1 Individuals with at least 50 landings in colour training (and adifference in at least one landing between the first and last ten landings) were


advanced to shape training Individuals were included in the analysis of shapelearning rate if they had at least 20 landings during shape training There was nodifference between treatment groups in total number of shape-training landings(F312 = 033 P = 080) Individuals were included in the analysis of target landingrate if there were at least two searching sessions during shape training for whichlanding rate could be calculated

RESULTS (2)

Learning during colour training

Overall during colour training females learned to choose the rewarding target overthe unrewarding target as their final error rate was significantly lower then theirinitial error rate (t22 = 383 P = 00009) There was no significant differencebetween the four colour training groups in training performance (initial error-finalerror F319 = 131 P = 031) or in initial error (F319 = 225 P = 012) althoughthere was a trend for initial error to be highest in the green against green treatmentand lowest in the blue against brown treatment

Learning of shape

We measured over the course of shape training changes in frequency of landingon the rewarding novel shape (12-prong versus three-prong target) In contrast tolearning target colour individuals did not learn overall to discriminate between thetwo shapes because the difference between their initial and final errors was notsignificantly greater than zero (mean (SE) = 0053 (004) t15 = minus13 P = 020)Rather changes in response to shape depended on treatment Individuals colourtrained to green targets against a green background improved significantly more inshape discrimination than did individuals colour trained to green targets on a brownbackground (fig 5 F18 = 593 P = 0041) In contrast individuals colour trainedto blue targets against a green background improved less in shape discriminationthan individuals colour trained to blue targets on a brown background however thisdifference was not statistically significant (fig 5 F14 = 384 P = 012) Takingthese results together colour training on a target colourbackground colour appearedto facilitate shape learning on the same target colourbackground combination

Target landing rate during shape training

The interaction between target colour and treatment group on shape discriminationwas paralleled by an interaction in terms of landing rate on each target duringthe shape training session (fig 5) Individuals colour trained to green targetsagainst a green background enjoyed significantly higher target landing rates inthe shape training phase than individuals trained to green targets against a brownbackground (fig 5 P = 003) In contrast individuals colour trained to blue against


a green background had target landing rates in the shape training phase that werevirtually identical to individuals colour trained to blue against a brown background(fig 5 P = 076) While treatment group alone (colour training background) wasmarginally significant in explaining variation in target landing rates during the shapetraining (F118 = 419 P = 0055) the overall statistical model for landing rate didnot identify training target colour treatment or the interaction as significant effectspossibly due to low power (background colour F116 = 304 P = 010 targetcolour F116 = 008 P = 078 target times background F116 = 164 P = 022)In conclusion while sample size is limited results suggest that experience witha green background increases target landing rate in a novel task against a greenbackground when butterflies are colour trained on green targets but not blueones

Learning to discriminate against background

We tested whether females learned to avoid landing on the background colourduring colour training and whether this discrimination was remembered duringshape training Overall individuals significantly improved their ability to avoidlanding on the background during colour training (t24 = 331 P = 0003)decreasing the estimated probability of landing on the background almost three-fold on average from 030 (SE = 0047) to 011 (0025) However individuals didnot retain this ability to avoid the background between colour training and shapetraining For each individual we used logistic regression to estimate the initialprobability of landing on background during colour training we used a separatelogistic regression to estimate the initial probability of landing on backgroundduring shape training We found that the initial background landing error duringshape training was equal to or higher than the initial error during colour training(fig 5)

The difference in frequency of background landing errors between colour trainingand shape training sessions depended on target colour In a full statistical modeltreatment group (experience with shape-training background colour) had no effecton changes in background landing frequency between colour and shape trainingwhile target colour was marginally significant individuals colour trained to bluetargets had higher error rates during shape training than colour training (effects ondifference between initial error in colour training and error during shape trainingtarget colour F112 = 444 P = 0056 colour training background F112 = 294P = 011 interaction NS) In summary our analysis of responses to backgroundsuggests that i) females learn to discriminate against the background ii) thisdiscrimination ability is not remembered in a new context and iii) increases inbackground landing mistakes during a novel task are especially high for individualslacking experience with the background colour of the novel task (either green targetsor a green background)


DISCUSSION

This research explored the relevance of variation in sensory environments toforaging behaviour i) are cues learned independent of the sensory environmentand ii) are features of the sensory environment learned and used to advantage inlearning novel tasks We address each question in turn in the next two sections

Are cues learned independently of the sensory environment

For host-searching butterflies the answer appears to be lsquonorsquo In this study cue learn-ing depended on the sensory environment in which cues are experienced Whenfemale butterflies were trained to either red or green targets against a brown back-ground their performance on the colour task was worse when tested subsequentlyon a green background relative to control individuals tested subsequently on thesame brown background (fig 4) In contrast for females trained against a greenbackground there was no effect of switching to a brown test background

These results suggest that butterflies learn characteristics of a green backgroundChanges in the relative conspicuousness of the rewarding targets (fig 1) cannotexplain these results because the pattern held for both red and green rewardingtargets (fig 4) That a change in background signals a new context (Lotto andChittka 2005) and the irrelevance of previously-learned associations cannotexplain the results either because changes to green but not to brown backgroundsresulted in less (or no) retention of learned associations (fig 4)

Several non-mutually exclusive mechanisms can explain why learned cues canbe extrapolated from green to brown backgrounds but not from brown to greenones First whether or not females attend to the background during learning maydepend on the overall conspicuousness of both targets In training against a greenbackground the green target (sometimes S+ sometimes S0) appears somewhatcryptic as the hue (wavelength of peak reflectance) of the target closely matchesthat of the background (fig 1) Thus under cryptic conditions females may learnbackground characteristics permitting them to discriminate the green target from thegreen background whether the green target is rewarding or not In training againstbrown by contrast both targets appear to be more conspicuous (fig 1) For thisreason background characteristics may not be learned or learned as well for cuesagainst brown backgrounds

Alternatively the result may relate less to crypticity of targets and more to thefact that green is intrinsically highly stimulating to females engaged in host search(eg initial error rate is biased towards green fig 3) The added stimulation inthe green background may somehow disrupt discrimination between targets whenfemales have been trained on a non-stimulating brown background In this caseexperience with the green background permits females to exclude the backgroundas a candidate for host tissue A brown background on the other hand does notrepresent a potential host and consequently does not disrupt discrimination betweentargets Distinguishing between these two mechanisms would require manipulating


crypticity independently of intrinsic stimulation (for example adding a trainingtreatment using red targets against red background)

Are characteristics of the background learned and applied to novel tasks

The answer appears to be a qualified lsquoyesrsquo Experiment 2 provided evidence thatsome characteristics of the background are learned and applied to novel tasks butthat such learning is selective Training to a green colour against a green backgroundfacilitated improved performance in a novel shape discrimination task of the sametarget colour-background colour combination (fig 5) compared to individuals withtraining to a green colour against a brown background In contrast individualstrained to a blue colour showed no clear effect of training background on learning ofthe shape discrimination task (fig 5) We tested for a possible mechanism involvinglearning and remembering to avoid landing on the background While individualslearned to refrain from landing on the background during training to colour theywere apparently unable to retain this discrimination when transferred to a shapetraining task (fig 5) Hence the component of background learning that may betransferred to learning of a novel task involves something different than learning torefrain from landing on the green background

Several non-mutually exclusive mechanisms can explain why background learn-ing appears to be adopted under training to green against green but not under train-ing to blue against green First background learning may occur only under cryp-tic conditions However we note that in Experiment 2 green against green wasless cryptic than in Experiment 1 because the background was darker (see fig 1)Second background learning may occur only when the background colour is in-nately attractive (green) and must therefore be actively ignored during host searchAs above these hypotheses could be distinguished by manipulating crypticity inde-pendently of intrinsic stimulation (for instance adding a red against red treatmentcondition in the first training phase) Finally because perceptual learning is oftenhighly specific (reviewed by Sathian 1998) background learning may only be ex-trapolated to novel tasks when characteristics of both the target and the backgroundare similar between tasks (ie green targets and green backgrounds fig 5)

Perceptual learning and sensory environments

Perceptual learning involves learning to detect objects of interest be they fooditems in foraging or conspecifics in communication The present study suggests thatperceptual learning is dependent on sensory environment (fig 4) and that learningcharacteristics of the background occurs simultaneously with the learning of cues(fig 5) Our results are broadly applicable to other systems in which perceptuallearning has been studied For example search image formation as in birds learningto detect cryptic prey (Pietrewicz and Kamil 1979) is considered to be a kindof perceptual learning Traditionally discussion of search image formation hasemphasised the learning of features of the prey per se Our results suggest that search


image formation may involve learning of prey cues and learning of backgroundfeatures as suggested by some theory and empirical evidence in neuroscience andpsychology (Dosher and Lu 1998 2000 Sigala and Logothetis 2002 Gold et al2004 Yang and Maunsell 2004)

If search image formation involves learning both of prey and of background ourperspectives on prey strategies to defeat search image formation may need to bebroadened For example it has been proposed that cryptic prey benefit by beingvariable in phenotype (Bond and Kamil 2002) Our present results if generalisablesuggest that prey might also benefit by occurring against varying backgrounds

This research also has relevance for a phenomenon in the search image literatureknown as background cuing When cryptic signals are associated with certainbackground environments animals appear to use the background to prime a searchimage for the associated signal this form of learning is called background cuing(Kono et al 1998) The results of this study suggest that there may be inherentmechanisms in perceptual learning to account for background cuing the sensoryenvironment appears to be learned in association with cues especially when cuesare cryptic (figs 4 5)

Our perspective on the value of learning from the standpoint of the predator orthe herbivore may need to be broadened as well For instance it is well knownthat the learning of one task can facilitate learning of novel tasks (Shettleworth1998) Our results suggest that one means by which such facilitation can occur isthrough learning of background characteristics which are transferred to other tasksSuch extrapolation may be relevant to insect learning in nature For instance a beeor butterflyrsquos learning to ignore the background in foraging for one flower type(Goulson 2000) may facilitate learning subsequently of another flower type in thesame or similar sensory environment

Future directions

In the Battus-Aristolochia system in southern Arizona we are only beginning tounderstand the nature of sensory variation in the habitat in relation to variation invisual host plant cues In nature (and in contrast to our experimental conditions)both green forms and red forms of the host A watsoni are generally highly crypticto the human observer albeit for different reasons Green forms are cryptic againstgreen foliage a crypticity which increases markedly after summer monsoon rainsin contrast red forms are so dark as to lsquohidersquo among the shadows of vegetationsoil and rock What is known about papilionid vision (reviewed by Arikawa 2003)suggests that Battus females must also cope with some degree of visual noise foreach colour form If so it may benefit females to learn not only the various coloursof host tissue but also elements of the background against which each colour formoccurs Certainly anyone who watches females engaged in host search in the fieldgains an immediate sense that females would benefit by learning features of thebackground This is because females identify host plant in part by tasting the leafsurface with chemoreceptors on their foretarsi In the field host-searching females


alight briefly but frequently on objects that are not Aristolochia plants includingmainly non-host vegetation but also dead twigs dirt and stones In fact suchlsquomistakenrsquo landings which must consume time and may increase vulnerability toground-borne predators are far more numerous than landings on actual hosts (by asmuch as two orders of magnitude D Papaj unpubl)

Apart from understanding the relevance of this study for this particular insect-host interaction the study needs to be repeated in other systems and with largersample sizes Testing multiple sets of cryptic targets and background conditionsmay clarify the mechanisms underlying background learning To what extent isthe sensory environment dealt with through associative learning versus alternativeprocesses such as sensory adaptation and habituation Sensory adaptation andhabituation which involve reduced responses to recurrent stimuli at the level ofsensory receptors and neural circuits respectively (Torre et al 1995 Dalton2000) are considered to be strategies for coping with noise (Torre et al 1995Pierce et al 1995) sometimes exerting long-lasting effects (Dalton and Wysocki1996 Fischer et al 2000 Bee and Gerhardt 2001 Rose and Rankin 2001Simonds and Plowright 2004) It is likely that learning backgrounds reflects acombination of sensory adaptation habituation and associative learning of featuresof unrewarding stimuli but the relative importance of the processes may vary fromone circumstance to the next and from one species to the next

Apart from the mechanisms of background learning the results raise many otherquestions For instance if colour learning is not independent of background colourhow does the phenomenon of colour constancy develop in Lepidoptera (Kinoshitaand Arikawa 2000) Does colour constancy itself involve the integration of learningtarget colour and learning background colour What constitutes noise in varioussensory modalities and does noise in one modality tend to be more constrainingthan in other modalities Are there innate biases in terms of coping with sensorybackgrounds Given that perceptual learning is often highly specific (reviewed bySathian 1998) what determines the extent to which background learning can beapplied to novel tasks Ecological and behavioural research should consider theconsequences of signals and cues being embedded in a sensory environment

ACKNOWLEDGEMENTS

Thanks to Natasha Pearce and Kirsten Selheim for assistance in data collection andto Ingrid Lindstrom Josh Garcia and Brad Worden for help in insect rearing Weare grateful to members of the Papaj lab for feedback on an earlier incarnation ofthis work Emilie Snell-Rood was supported by an NSF predoctoral fellowship Thiswork was supported by an NSF-IBN grant to Daniel Papaj

REFERENCES

Allard RA amp Papaj DR (1996) Learning of leaf shape by pipevine swallowtail butterflies A testusing artificial leaf models J Insect Behav 9 961-967


Arikawa K (2003) Spectral organization of the eye of a butterfly Papilio J Comp Physiol A 189791-800

Bee MA amp Gerhardt HC (2001) Habituation as a mechanism of reduced aggression betweenneighboring territorial male bullfrogs (Rana catesbeiana) J Comp Psychol 115 68-82

Bond AB amp Kamil AC (2002) Visual predators select for crypticity and polymorphism in virtualprey Nature 415 609-613

Brumm H (2004) The impact of environmental noise on song amplitude in a territorial bird J AnimEcol 73 434-440

Brumm H amp Todt D (2002) Noise-dependent song amplitude regulation in a territorial songbirdAnim Behav 63 891-897

Cunningham J Nicol T King C Zecker SG amp Kraus N (2002) Effects of noise and cueenhancement on neural responses to speech in auditory midbrain thalamus and cortex HearRes 169 97-111

Cynx J Lewis R Tavil B amp Tse H (1998) Amplitude regulation of vocalizations in noise by asongbird Taeniopygia guttata Anim Behav 56 107-113

Dalton P amp Wysocki CJ (1996) The nature and duration of adaptation following long-term odorexposure Percept Psychophys 58 781-792

Dalton P (2000) Psychophysical and behavioral characteristics of olfactory adaptation Chem Senses25 487-492

Dosher BA amp Lu Z-L (1998) Perceptual learning reflects external noise filtering and internal noisereduction through channel reweighting Proc Natl Acad Sci USA 95 13988-13993

Dosher BA amp Lu Z-L (2000) Mechanisms of perceptual attention in precuing of location VisionRes 40 1269-1292

Endler JA (1992) Signals signal conditions and the direction of evolution Am Nat 139 S125-S153

Endler JA (1993) The color of light in forests and its implications Ecol Monogr 63 1-27Endler JA amp Theacutery M (1996) Interacting effects of lek placement display behavior ambient light

and colour patterns in three neotropical forest-dwelling birds Am Nat 148 421-458Fischer TM Yuan JW amp Carew TJ (2000) Dynamic regulation of the siphon withdrawal reflex

of Aplysia californica in response to changes in the ambient tactile environment Behav Neurosci114 1209-1222

Gold JM Sekuler AB amp Bennett PJ (2004) Characterizing perceptual learning with externalnoise Cogn Sci 28 167-207

Goldstone RL (1998) Perceptual learning Annu Rev Psychol 49 585-612Goulson D (2000) Are insects flower constant because they use search images to find flowers Oikos

88 547-552Kinoshita M amp Arikawa K (2000) Color constancy of the swallowtail butterfly Papilio xuthus J

Exp Biol 203 3521-3530Kono H Reid PJ amp Kamil AC (1998) The effect of background cuing on prey detection Anim

Behav 56 963-972Langley CM (1996) Search images selective attention to specific visual features of prey J Exp

Psychol Anim Behav Process 22 152-163Leal M amp Fleishman LJ (2004) Differences in visual signal design and detectability between

allopatric populations of Anolis lizards Am Nat 163 26-39Lengagne T amp Slater PJB (2002) The effects of rain on acoustic communication tawny owls have

good reason for calling less in wet weather Proc R Soc Lond B 269 2121-2125Lengagne T Aubin T Lauga J amp Jouventin P (1999) How do king penguins (Aptenodytes

patagonicus) apply the mathematical theory of information to communicate in windy conditionsProc R Soc Lond B 266 1623-1628

Lotto RB amp Chittka L (2005) Seeing the light illumination as a contextual cue to color choicebehavior in bumblebees Proc Natl Acad Sci USA 102 3852-3856


Lynn SK Cnaani J amp Papaj DR (2005) Peak shift discrimination learning as a mechanism ofsignal evolution Evolution 59 1300-1305

Marchetti K (1993) Dark habitats and bright birds illustrate the role of the environment in speciesdivergence Nature 11 149-152

Morton EA (1975) Ecological sources of selection on avian sounds Am Nat 109 17-34Naguib M (2003) Reverberation of rapid and slow trills implications for signal adaptations to long-

range communication J Acoust Soc Am 113 749-1756Papaj DR (1986) Interpopulation differences in host preference and the evolution of learning in the

butterfly Battus philenor Evolution 40 518-530Papaj DR amp Prokopy RJ (1989) Ecological and evolutionary aspects of learning in phytophagous

insects Annu Rev Entomol 34 315-350Pietrewicz AT amp Kamil AC (1979) Search image formation in the blue jay (Cyanocitta cristata)

Science 204 1332-1333Plaisted KC amp Mackintosh NJ (1995) Visual search for cryptic stimuli in pigeons implications

for the search image and search rate hypothesis Anim Behav 50 1219-1232Richards DG amp Wiley RH (1980) Reverberations and amplitude fluctuations in the propagation of

sound in a forest implications for animal communication Am Nat 115 381-399Rose JK amp Rankin CH (2001) Analyses of habituation in Caenorhabditis elegans Learn Mem

8 63-69Ryan MJ amp Brenowitz EA (1985) The role of body size phylogeny and ambient noise in the

evolution of bird song Am Nat 126 87-100Sathian K (1998) Perceptual learning Curr Sci 75 451-457Shettleworth SJ (1998) Cognition Evolution and Behavior New York Oxford University PressSigala N amp Logothetis NK (2002) Visual categorization shapes feature selectivity in the primate

temporal cortex Nature 415 318-320Simonds V amp Plowright CMS (2004) How do bumblebees first find flowers Unlearned approach

responses and habituation Anim Behav 67 379-386Slabbekoorn H amp Smith TB (2002) Habitat-dependent song divergence in the little greenbul an

analysis of environmental selection pressures on acoustic signals Evolution 56 1849-1858Torre V Ashmore JF Lamb TD amp Menini A (1995) Transduction and adaptation in sensory

receptor cells J Neurosci 15 7757-7768Vaina LM Sundareswaran V amp Harris JG (1995) Learning to ignore psychophysics and

computational modeling of fast learning of direction in noisy motion stimuli Cogn Brain Res2 155-163

Van Staaden MJ amp Roumlmer H (1997) Sexual signalling in bladder grasshoppers tactical design formaximizing calling range J Exp Biol 200 2597-2608

Visser JH (1986) Host odor perception in phytophagous insects Annu Rev Entomol 31 121-144Watanabe T Naacutentildeez JE amp Sasaki Y (2001) Perceptual learning without perception Nature 413

844-848Wehner R (1987) lsquoMatched filtersrsquo ndash neural models of the external world J Comp Physiol A 161

511-531Weiss MR amp Papaj DR (2003) Butterfly color learning in two different behavioral contexts How

much can a butterfly keep in mind Anim Behav 65 425-434Wiley RH (1994) Errors exaggeration and deception in animal communication In LA Real (Ed)

Behavioral Mechanisms in Evolutionary Ecology pp 157-189 Chicago University of ChicagoPress

Yang T amp Maunsell JHR (2004) The effect of perceptual learning on neuronal responses in monkeyvisual area V4 J Neurosci 24 1617-1626


termed lsquonoisersquo which are similar to and thereby often obscure stimuli of inter-est Sensory environments are moreover inherently variable over time and spaceFor example different habitats are characterised by visual environments that varyin terms of the spectral composition of ambient light and the reflectance proper-ties of vegetation and substrates (Endler 1993 Marchetti 1993 Endler and Theacutery1996 Leal and Fleishman 2004) Ambient sound levels also vary significantly be-tween habitat types some habitats being consistently noisy over certain frequencyranges (eg Morton 1975 Ryan and Brenowitz 1985 Slabbekoorn and Smith2002) Acoustic noise due to sound reverberation varies between seasons in de-ciduous forests (Naguib 2003) and between habitats (Richards and Wiley 1980)Phytophagous insects must discriminate host plants against olfactory cues from non-host plants olfactory background characteristics depend on factors such as non-hostplant species present vegetation structure weather conditions and in an agricul-tural context the diversity and spacing of crops (reviewed by Visser 1986)

Species often adapt to cope with characteristics specific to particular sensoryenvironments For example signals used in communication evolve to travel ef-ficiently through particular habitats (eg Ryan and Brenowitz 1985 Marchetti1993 Endler and Theacutery 1996 Slabbekoorn and Smith 2002 Cunningham et al2002 reviewed by Wiley 1994) Sensory systems evolve not only to perceive rele-vant information (Wehner 1987) but also to filter out irrelevant background infor-mation (reviewed by Endler 1992)

Additionally individuals may dynamically adjust signals to spatiotemporal varia-tion in sensory environments Such individual level responses have been best stud-ied from the perspective of the signaller in communication systems When signalsare obscured by the background signals tend to increase in intensity or in numberof repetitive elements (Cynx et al 1998 Lengagne et al 1999 Brumm and Todt2002 Brumm 2004) Alternatively signallers may opt to signal later when back-ground conditions change (eg delays in communication due to weather conditions(Lengagne and Slater 2002) or time of day (van Staaden and Roumlmer 1997))

Individual-level plasticity in response to sensory environments has been muchless studied from the standpoint of a signal receiver However a receiver may alsorespond to variation in the sensory environment whether with respect to signalsfrom conspecifics or cues from predators or prey Receivers could cope withvariability in sensory environments in several non-mutually exclusive ways Forinstance signal detection theory predicts that receivers may adjust their thresholdof response to a stimulus of interest so as to simultaneously maximise responsesto the relevant stimulus and minimise responses to similar but irrelevant stimuli(so-called lsquofalse alarmsrsquo) A signal detection framework (cf Wiley 1994) hasrecently been applied with success to analysis of learning of floral cues by foragingbumblebees (Lynn Cnaani and Papaj 2005)

Alternatively individuals experiencing a particular sensory environment maylearn through a process of perceptual learning to ignore features of the environmentand thus improve subsequent performance (Watanabe et al 2001) Perceptual




METHODS

Study system















Learning assays




Data analysis


y = 1

1 + eminus(a+bx)(1)






Experimental design



Learning criteria


RESULTS (1)












Experimental design



Fig

ure

5Pe

rfor

man

ceon

ano

veld

iscr

imin

atio

nta

sk(s

hape

)fol

low

ing

expe

rien

cew

ithpa

rtic

ular

back

grou

nds

duri

ngco

lour

trai

ning

The

Y-a

xis

repr

esen

tsa

part

icul

arm

easu

reof

perf

orm

ance

duri

ngsh

ape

trai

ning

The

X-a

xis

isas

infig

ure

3ex

cept

lsquoBrsquor

epre

sent

sbl

ueta

rget

san

dsh

apes

inth

elo

wer

box

repr

esen

tth

etw

ota

rget

sus

edin

the

shap

etr

aini

ngT

hele

ft-h

and

grap

hsh

ows

chan

ges

inth

eco

rrec

tch

oice

ofth

ere

war

ding

shap

eov

erth

esh

ape

trai

ning

sess

ion

(ie

di

ffer

ence

betw

een

initi

alan

dfin

aler

rors

see

fig2

)th

usm

ore

posi

tive

valu

esre

pres

ent

impr

ovem

ents

indi

scri

min

atio

ndu

ring

shap

etr

aini

ngT

hem

iddl

egr

aph

show

sla

ndin

gra

tes

onta

rget

sdu

ring

shap

etr

aini

ng(n

ote

log

scal

e)t

hus

mor

epo

sitiv

eva

lues

repr

esen

thi

gher

perf

orm

ance

For

thes

efir

sttw

om

easu

res

ofpe

rfor

man

cet

hegr

oup

with

prev

ious

expe

rien

ceon

the

shap

e-tr

aini

ngba

ckgr

ound

(con

trol

grou

p)pe

rfor

med

bette

rth

anth

egr

oup

wit

hout

such

expe

rien

ce

only

whe

nco

lour

trai

ning

occu

rred

ongr

een

targ

ets

The

righ

t-ha

ndgr

aph

show

sch

ange

sin

abili

tyto

igno

reth

eba

ckgr

ound

betw

een

colo

urtr

aini

ngan

dsh

ape

trai

ning

whi

lein

divi

dual

sle

arne

dto

igno

reth

eba

ckgr

ound

duri

ngco

lour

trai

ning

(see

text

)th

isab

ility

was

notr

etai

ned

insh

ape

trai

ning

asth

edi

ffer

ence

betw

een

initi

alba

ckgr

ound

land

ing

duri

ngco

lour

trai

ning

and

shap

etr

aini

ngw

asle

ssth

anor

equa

lto

zero

(sig

nify

ing

noch

ange

or

anin

crea

sein

back

grou

ndla

ndin

gsbe

twee

ntr

aini

ngse

ssio

nsr

espe

ctiv

ely)







Learning criteria




RESULTS (2)



Learning of shape










DISCUSSION



















Future directions






ACKNOWLEDGEMENTS


REFERENCES























































METHODS

Study system















Learning assays




Data analysis


y = 1

1 + eminus(a+bx)(1)






Experimental design



Learning criteria


RESULTS (1)












Experimental design



Fig

ure

5Pe

rfor

man

ceon

ano

veld

iscr

imin

atio

nta

sk(s

hape

)fol

low

ing

expe

rien

cew

ithpa

rtic

ular

back

grou

nds

duri

ngco

lour

trai

ning

The

Y-a

xis

repr

esen

tsa

part

icul

arm

easu

reof

perf

orm

ance

duri

ngsh

ape

trai

ning

The

X-a

xis

isas

infig

ure

3ex

cept

lsquoBrsquor

epre

sent

sbl

ueta

rget

san

dsh

apes

inth

elo

wer

box

repr

esen

tth

etw

ota

rget

sus

edin

the

shap

etr

aini

ngT

hele

ft-h

and

grap

hsh

ows

chan

ges

inth

eco

rrec

tch

oice

ofth

ere

war

ding

shap

eov

erth

esh

ape

trai

ning

sess

ion

(ie

di

ffer

ence

betw

een

initi

alan

dfin

aler

rors

see

fig2

)th

usm

ore

posi

tive

valu

esre

pres

ent

impr

ovem

ents

indi

scri

min

atio

ndu

ring

shap

etr

aini

ngT

hem

iddl

egr

aph

show

sla

ndin

gra

tes

onta

rget

sdu

ring

shap

etr

aini

ng(n

ote

log

scal

e)t

hus

mor

epo

sitiv

eva

lues

repr

esen

thi

gher

perf

orm

ance

For

thes

efir

sttw

om

easu

res

ofpe

rfor

man

cet

hegr

oup

with

prev

ious

expe

rien

ceon

the

shap

e-tr

aini

ngba

ckgr

ound

(con

trol

grou

p)pe

rfor

med

bette

rth

anth

egr

oup

wit

hout

such

expe

rien

ce

only

whe

nco

lour

trai

ning

occu

rred

ongr

een

targ

ets

The

righ

t-ha

ndgr

aph

show

sch

ange

sin

abili

tyto

igno

reth

eba

ckgr

ound

betw

een

colo

urtr

aini

ngan

dsh

ape

trai

ning

whi

lein

divi

dual

sle

arne

dto

igno

reth

eba

ckgr

ound

duri

ngco

lour

trai

ning

(see

text

)th

isab

ility

was

notr

etai

ned

insh

ape

trai

ning

asth

edi

ffer

ence

betw

een

initi

alba

ckgr

ound

land

ing

duri

ngco

lour

trai

ning

and

shap

etr

aini

ngw

asle

ssth

anor

equa

lto

zero

(sig

nify

ing

noch

ange

or

anin

crea

sein

back

grou

ndla

ndin

gsbe

twee

ntr

aini

ngse

ssio

nsr

espe

ctiv

ely)







Learning criteria




RESULTS (2)



Learning of shape










DISCUSSION



















Future directions






ACKNOWLEDGEMENTS


REFERENCES

































































Learning assays




Data analysis


y = 1

1 + eminus(a+bx)(1)






Experimental design



Learning criteria


RESULTS (1)












Experimental design



Fig

ure

5Pe

rfor

man

ceon

ano

veld

iscr

imin

atio

nta

sk(s

hape

)fol

low

ing

expe

rien

cew

ithpa

rtic

ular

back

grou

nds

duri

ngco

lour

trai

ning

The

Y-a

xis

repr

esen

tsa

part

icul

arm

easu

reof

perf

orm

ance

duri

ngsh

ape

trai

ning

The

X-a

xis

isas

infig

ure

3ex

cept

lsquoBrsquor

epre

sent

sbl

ueta

rget

san

dsh

apes

inth

elo

wer

box

repr

esen

tth

etw

ota

rget

sus

edin

the

shap

etr

aini

ngT

hele

ft-h

and

grap

hsh

ows

chan

ges

inth

eco

rrec

tch

oice

ofth

ere

war

ding

shap

eov

erth

esh

ape

trai

ning

sess

ion

(ie

di

ffer

ence

betw

een

initi

alan

dfin

aler

rors

see

fig2

)th

usm

ore

posi

tive

valu

esre

pres

ent

impr

ovem

ents

indi

scri

min

atio

ndu

ring

shap

etr

aini

ngT

hem

iddl

egr

aph

show

sla

ndin

gra

tes

onta

rget

sdu

ring

shap

etr

aini

ng(n

ote

log

scal

e)t

hus

mor

epo

sitiv

eva

lues

repr

esen

thi

gher

perf

orm

ance

For

thes

efir

sttw

om

easu

res

ofpe

rfor

man

cet

hegr

oup

with

prev

ious

expe

rien

ceon

the

shap

e-tr

aini

ngba

ckgr

ound

(con

trol

grou

p)pe

rfor

med

bette

rth

anth

egr

oup

wit

hout

such

expe

rien

ce

only

whe

nco

lour

trai

ning

occu

rred

ongr

een

targ

ets

The

righ

t-ha

ndgr

aph

show

sch

ange

sin

abili

tyto

igno

reth

eba

ckgr

ound

betw

een

colo

urtr

aini

ngan

dsh

ape

trai

ning

whi

lein

divi

dual

sle

arne

dto

igno

reth

eba

ckgr

ound

duri

ngco

lour

trai

ning

(see

text

)th

isab

ility

was

notr

etai

ned

insh

ape

trai

ning

asth

edi

ffer

ence

betw

een

initi

alba

ckgr

ound

land

ing

duri

ngco

lour

trai

ning

and

shap

etr

aini

ngw

asle

ssth

anor

equa

lto

zero

(sig

nify

ing

noch

ange

or

anin

crea

sein

back

grou

ndla

ndin

gsbe

twee

ntr

aini

ngse

ssio

nsr

espe

ctiv

ely)







Learning criteria




RESULTS (2)



Learning of shape










DISCUSSION



















Future directions






ACKNOWLEDGEMENTS


REFERENCES
























































Experimental design



Learning criteria


RESULTS (1)












Experimental design



Fig

ure

5Pe

rfor

man

ceon

ano

veld

iscr

imin

atio

nta

sk(s

hape

)fol

low

ing

expe

rien

cew

ithpa

rtic

ular

back

grou

nds

duri

ngco

lour

trai

ning

The

Y-a

xis

repr

esen

tsa

part

icul

arm

easu

reof

perf

orm

ance

duri

ngsh

ape

trai

ning

The

X-a

xis

isas

infig

ure

3ex

cept

lsquoBrsquor

epre

sent

sbl

ueta

rget

san

dsh

apes

inth

elo

wer

box

repr

esen

tth

etw

ota

rget

sus

edin

the

shap

etr

aini

ngT

hele

ft-h

and

grap

hsh

ows

chan

ges

inth

eco

rrec

tch

oice

ofth

ere

war

ding

shap

eov

erth

esh

ape

trai

ning

sess

ion

(ie

di

ffer

ence

betw

een

initi

alan

dfin

aler

rors

see

fig2

)th

usm

ore

posi

tive

valu

esre

pres

ent

impr

ovem

ents

indi

scri

min

atio

ndu

ring

shap

etr

aini

ngT

hem

iddl

egr

aph

show

sla

ndin

gra

tes

onta

rget

sdu

ring

shap

etr

aini

ng(n

ote

log

scal

e)t

hus

mor

epo

sitiv

eva

lues

repr

esen

thi

gher

perf

orm

ance

For

thes

efir

sttw

om

easu

res

ofpe

rfor

man

cet

hegr

oup

with

prev

ious

expe

rien

ceon

the

shap

e-tr

aini

ngba

ckgr

ound

(con

trol

grou

p)pe

rfor

med

bette

rth

anth

egr

oup

wit

hout

such

expe

rien

ce

only

whe

nco

lour

trai

ning

occu

rred

ongr

een

targ

ets

The

righ

t-ha

ndgr

aph

show

sch

ange

sin

abili

tyto

igno

reth

eba

ckgr

ound

betw

een

colo

urtr

aini

ngan

dsh

ape

trai

ning

whi

lein

divi

dual

sle

arne

dto

igno

reth

eba

ckgr

ound

duri

ngco

lour

trai

ning

(see

text

)th

isab

ility

was

notr

etai

ned

insh

ape

trai

ning

asth

edi

ffer

ence

betw

een

initi

alba

ckgr

ound

land

ing

duri

ngco

lour

trai

ning

and

shap

etr

aini

ngw

asle

ssth

anor

equa

lto

zero

(sig

nify

ing

noch

ange

or

anin

crea

sein

back

grou

ndla

ndin

gsbe

twee

ntr

aini

ngse

ssio

nsr

espe

ctiv

ely)







Learning criteria




RESULTS (2)



Learning of shape










DISCUSSION



















Future directions






ACKNOWLEDGEMENTS


REFERENCES





















































Learning criteria


RESULTS (1)












Experimental design



Fig

ure

5Pe

rfor

man

ceon

ano

veld

iscr

imin

atio

nta

sk(s

hape

)fol

low

ing

expe

rien

cew

ithpa

rtic

ular

back

grou

nds

duri

ngco

lour

trai

ning

The

Y-a

xis

repr

esen

tsa

part

icul

arm

easu

reof

perf

orm

ance

duri

ngsh

ape

trai

ning

The

X-a

xis

isas

infig

ure

3ex

cept

lsquoBrsquor

epre

sent

sbl

ueta

rget

san

dsh

apes

inth

elo

wer

box

repr

esen

tth

etw

ota

rget

sus

edin

the

shap

etr

aini

ngT

hele

ft-h

and

grap

hsh

ows

chan

ges

inth

eco

rrec

tch

oice

ofth

ere

war

ding

shap

eov

erth

esh

ape

trai

ning

sess

ion

(ie

di

ffer

ence

betw

een

initi

alan

dfin

aler

rors

see

fig2

)th

usm

ore

posi

tive

valu

esre

pres

ent

impr

ovem

ents

indi

scri

min

atio

ndu

ring

shap

etr

aini

ngT

hem

iddl

egr

aph

show

sla

ndin

gra

tes

onta

rget

sdu

ring

shap

etr

aini

ng(n

ote

log

scal

e)t

hus

mor

epo

sitiv

eva

lues

repr

esen

thi

gher

perf

orm

ance

For

thes

efir

sttw

om

easu

res

ofpe

rfor

man

cet

hegr

oup

with

prev

ious

expe

rien

ceon

the

shap

e-tr

aini

ngba

ckgr

ound

(con

trol

grou

p)pe

rfor

med

bette

rth

anth

egr

oup

wit

hout

such

expe

rien

ce

only

whe

nco

lour

trai

ning

occu

rred

ongr

een

targ

ets

The

righ

t-ha

ndgr

aph

show

sch

ange

sin

abili

tyto

igno

reth

eba

ckgr

ound

betw

een

colo

urtr

aini

ngan

dsh

ape

trai

ning

whi

lein

divi

dual

sle

arne

dto

igno

reth

eba

ckgr

ound

duri

ngco

lour

trai

ning

(see

text

)th

isab

ility

was

notr

etai

ned

insh

ape

trai

ning

asth

edi

ffer

ence

betw

een

initi

alba

ckgr

ound

land

ing

duri

ngco

lour

trai

ning

and

shap

etr

aini

ngw

asle

ssth

anor

equa

lto

zero

(sig

nify

ing

noch

ange

or

anin

crea

sein

back

grou

ndla

ndin

gsbe

twee

ntr

aini

ngse

ssio

nsr

espe

ctiv

ely)







Learning criteria




RESULTS (2)



Learning of shape










DISCUSSION



















Future directions






ACKNOWLEDGEMENTS


REFERENCES




























































Experimental design



Fig

ure

5Pe

rfor

man

ceon

ano

veld

iscr

imin

atio

nta

sk(s

hape

)fol

low

ing

expe

rien

cew

ithpa

rtic

ular

back

grou

nds

duri

ngco

lour

trai

ning

The

Y-a

xis

repr

esen

tsa

part

icul

arm

easu

reof

perf

orm

ance

duri

ngsh

ape

trai

ning

The

X-a

xis

isas

infig

ure

3ex

cept

lsquoBrsquor

epre

sent

sbl

ueta

rget

san

dsh

apes

inth

elo

wer

box

repr

esen

tth

etw

ota

rget

sus

edin

the

shap

etr

aini

ngT

hele

ft-h

and

grap

hsh

ows

chan

ges

inth

eco

rrec

tch

oice

ofth

ere

war

ding

shap

eov

erth

esh

ape

trai

ning

sess

ion

(ie

di

ffer

ence

betw

een

initi

alan

dfin

aler

rors

see

fig2

)th

usm

ore

posi

tive

valu

esre

pres

ent

impr

ovem

ents

indi

scri

min

atio

ndu

ring

shap

etr

aini

ngT

hem

iddl

egr

aph

show

sla

ndin

gra

tes

onta

rget

sdu

ring

shap

etr

aini

ng(n

ote

log

scal

e)t

hus

mor

epo

sitiv

eva

lues

repr

esen

thi

gher

perf

orm

ance

For

thes

efir

sttw

om

easu

res

ofpe

rfor

man

cet

hegr

oup

with

prev

ious

expe

rien

ceon

the

shap

e-tr

aini

ngba

ckgr

ound

(con

trol

grou

p)pe

rfor

med

bette

rth

anth

egr

oup

wit

hout

such

expe

rien

ce

only

whe

nco

lour

trai

ning

occu

rred

ongr

een

targ

ets

The

righ

t-ha

ndgr

aph

show

sch

ange

sin

abili

tyto

igno

reth

eba

ckgr

ound

betw

een

colo

urtr

aini

ngan

dsh

ape

trai

ning

whi

lein

divi

dual

sle

arne

dto

igno

reth

eba

ckgr

ound

duri

ngco

lour

trai

ning

(see

text

)th

isab

ility

was

notr

etai

ned

insh

ape

trai

ning

asth

edi

ffer

ence

betw

een

initi

alba

ckgr

ound

land

ing

duri

ngco

lour

trai

ning

and

shap

etr

aini

ngw

asle

ssth

anor

equa

lto

zero

(sig

nify

ing

noch

ange

or

anin

crea

sein

back

grou

ndla

ndin

gsbe

twee

ntr

aini

ngse

ssio

nsr

espe

ctiv

ely)







Learning criteria




RESULTS (2)



Learning of shape










DISCUSSION



















Future directions






ACKNOWLEDGEMENTS


REFERENCES





















































Fig

ure

5Pe

rfor

man

ceon

ano

veld

iscr

imin

atio

nta

sk(s

hape

)fol

low

ing

expe

rien

cew

ithpa

rtic

ular

back

grou

nds

duri

ngco

lour

trai

ning

The

Y-a

xis

repr

esen

tsa

part

icul

arm

easu

reof

perf

orm

ance

duri

ngsh

ape

trai

ning

The

X-a

xis

isas

infig

ure

3ex

cept

lsquoBrsquor

epre

sent

sbl

ueta

rget

san

dsh

apes

inth

elo

wer

box

repr

esen

tth

etw

ota

rget

sus

edin

the

shap

etr

aini

ngT

hele

ft-h

and

grap

hsh

ows

chan

ges

inth

eco

rrec

tch

oice

ofth

ere

war

ding

shap

eov

erth

esh

ape

trai

ning

sess

ion

(ie

di

ffer

ence

betw

een

initi

alan

dfin

aler

rors

see

fig2

)th

usm

ore

posi

tive

valu

esre

pres

ent

impr

ovem

ents

indi

scri

min

atio

ndu

ring

shap

etr

aini

ngT

hem

iddl

egr

aph

show

sla

ndin

gra

tes

onta

rget

sdu

ring

shap

etr

aini

ng(n

ote

log

scal

e)t

hus

mor

epo

sitiv

eva

lues

repr

esen

thi

gher

perf

orm

ance

For

thes

efir

sttw

om

easu

res

ofpe

rfor

man

cet

hegr

oup

with

prev

ious

expe

rien

ceon

the

shap

e-tr

aini

ngba

ckgr

ound

(con

trol

grou

p)pe

rfor

med

bette

rth

anth

egr

oup

wit

hout

such

expe

rien

ce

only

whe

nco

lour

trai

ning

occu

rred

ongr

een

targ

ets

The

righ

t-ha

ndgr

aph

show

sch

ange

sin

abili

tyto

igno

reth

eba

ckgr

ound

betw

een

colo

urtr

aini

ngan

dsh

ape

trai

ning

whi

lein

divi

dual

sle

arne

dto

igno

reth

eba

ckgr

ound

duri

ngco

lour

trai

ning

(see

text

)th

isab

ility

was

notr

etai

ned

insh

ape

trai

ning

asth

edi

ffer

ence

betw

een

initi

alba

ckgr

ound

land

ing

duri

ngco

lour

trai

ning

and

shap

etr

aini

ngw

asle

ssth

anor

equa

lto

zero

(sig

nify

ing

noch

ange

or

anin

crea

sein

back

grou

ndla

ndin

gsbe

twee

ntr

aini

ngse

ssio

nsr

espe

ctiv

ely)







Learning criteria




RESULTS (2)



Learning of shape










DISCUSSION



















Future directions






ACKNOWLEDGEMENTS


REFERENCES


























































Learning criteria




RESULTS (2)



Learning of shape










DISCUSSION



















Future directions






ACKNOWLEDGEMENTS


REFERENCES






















































RESULTS (2)



Learning of shape










DISCUSSION



















Future directions






ACKNOWLEDGEMENTS


REFERENCES


























































DISCUSSION



















Future directions






ACKNOWLEDGEMENTS


REFERENCES
































































Future directions






ACKNOWLEDGEMENTS


REFERENCES
























































ACKNOWLEDGEMENTS


REFERENCES




















































learning signals within sensory environments: does host cue … · 2019-03-29 · animal biology,...

Documents